The code is deterministic. If you are referencing timestamps and metadata it has nothing to do with program execution even though binary file is not the same.
Based on strong positive feedback from customers in our beta test program, @spacexai will make Grok 4.5 available to the public tomorrow.
It is an Opus-class model, but faster, more token-efficient and lower cost.
@_Evan_Boyle I have tried it in vscode as custom model i.e. headroom proxy. Checking the logs I did not see any improvement. But when I connected Hermes agent to headroom proxy I get 40 to 60% token savings. Still discovering daily new AI things and headroom is not to be dismissed IMHO.
@edzitron Also ran a Sonnet to make feature upgrade on medium sized app. It consumed 7% straight away. 1h... Then some minor prompts and was over 10% and it was mid day. I did not expect it to be this expensive maybe 2x but it is more like 10+x.
@mitchellh My experience is that Copilot profiler with Opus 4.8 successfully resolved slow computation issue in eval engine after i gave him bottleneck from aspire trace. The more you localize the issue the better result AI will give you. I had bad experience with agentic loops also.